Function create_knowledge_index

apps/knowledge/serializers/common.py:243–265

(knowledge_id=None, document_id=None)

def create_knowledge_index(knowledge_id=None, document_id=None):
    if knowledge_id is None and document_id is None:
        raise AppApiException(500, _('Knowledge ID or Document ID must be provided'))

    if knowledge_id is not None:
        k_id = knowledge_id
    else:
        document = QuerySet(Document).filter(id=document_id).first()
        k_id = document.knowledge_id

    sql = f"SELECT indexname, indexdef FROM pg_indexes WHERE tablename = 'embedding' AND indexname = 'embedding_hnsw_idx_{k_id}'"
    index = sql_execute(sql, [])
    if not index:
        sql = f"SELECT vector_dims(embedding) AS dims FROM embedding WHERE knowledge_id = '{k_id}' LIMIT 1"
        result = sql_execute(sql, [])
        if len(result) == 0:
            return
        dims = result[0]['dims']
        # 超过2000维度不创建索引，pgvector hnsw索引不支持超过2000维度
        if dims < 2000:
            sql = f"""CREATE INDEX "embedding_hnsw_idx_{k_id}" ON embedding USING hnsw ((embedding::vector({dims})) vector_cosine_ops) WHERE knowledge_id = '{k_id}'"""
            update_execute(sql, [])
            maxkb_logger.info(f'Created index for knowledge ID: {k_id}')


def drop_knowledge_index(knowledge_id=None, document_id=None):